neuron-request@HPLMS2.HPL.HP.COM ("Neuron-Digest Moderator Peter Marvit") (04/21/91)
Neuron Digest Saturday, 20 Apr 1991 Volume 7 : Issue 21 Today's Topics: Retina Simulator generalization power of NNs Re: Rigorous results on Fault Tolerance and Robustness (Are there any?) lack of generalization Neural Nets in Autonomous Land Vehicles Proceedings of Third NN and PDP Conference, Indiana-Purdue Neural Compuation Vol 3 Issue 1 TR available: Catastrophic forgetting prepint by Lumer & Huberman: "Binding Hierarchies:..." Neural Network Seminar Neural Nets Workshop (IWANN 91) Send submissions, questions, address maintenance and requests for old issues to "neuron-request@hplabs.hp.com" or "{any backbone,uunet}!hplabs!neuron-request" Use "ftp" to get old issues from hplpm.hpl.hp.com (15.255.176.205). ------------------------------------------------------------ Subject: Retina Simulator From: Robert Siminoff <siminoff@ifado.uucp> Date: Wed, 27 Mar 91 10:31:32 -0800 I can make available for the cost of duplicating and postage ($20) a computer simulation of the retina, written in Fortran 77. The model has been accepted for publication in BIOLOGICAL CYBERNETICS and a preprint of the paper can be provided. For more information send an E-mail message to siminoff@ifado.uucp. Robert Siminoff ------------------------------ Subject: generalization power of NNs From: phil neal <phil@iris.iphc.washington.edu> Date: Tue, 02 Apr 91 12:58:59 -0800 Dear NN people, I posted this to comp.ai.neural-nets as well. But I thought I might reach more/different people through this venue. I have a problem with the ability of a neural net to generalize. I have 600 observations of a 6 predictor variable input vector to classify these observations into 1 of 4 groups. I break the data into a 400 observation training set and a 200 observation test set. When I use a simple linear discriminant function with seperate covariance matrices and compare that against a NN with 6 input, 12 hidden and 4 output nodes. Here's what I get for correct classification rates: LDF NN train 48.5 59.0 test 42.0 37.0 And no matter how long I let the NN run, and no matter what number of hidden layer nodes, I always get about the same results. So, what's the deal ? Is my sample size too small ? Are there any good papers that cover this kind of problem ? I know I am violating the rule of thumb to have 10 times more training data than nodes in the net. But hey, data is expensive. Thanks, Phil Neal phil@iris.iphc.washington.edu direct to my workstation ------------------------------ Subject: Re: Rigorous results on Fault Tolerance and Robustness (Are there any?) From: mamisra%pollux.usc.edu@usc.edu (Manavendra Misra) Date: Tue, 02 Apr 91 14:52:09 -0800 L. Belfore at the Dept of Elec and Comp Engg, Marquette Univ, Milwaukee, WI 53233 and B. Johnson at the Dept of Elec Engg, Univ of Virginia, Charlottesville, VA 22901 are doing work in this field. Unfortunately, I do not have their email addresses. Manav. ------------------------------ Subject: lack of generalization From: phil neal <phil@iris.iphc.washington.edu> Date: Tue, 02 Apr 91 16:00:01 -0800 Dear NN people, This is a follow up to my posting on the lack of generalization of NNs. One cure for this problem that I have heard about is the following: I have heard of workers creating synthetic data from the data set they had. From what I understand, it goes something like this: 1. for each predictor variable in the training set a. Assume a distribution b. find the empirical parameters for that distribution using all the records in the training set. i.e a mean and s.d. on a normal or uniform 2. for as many passes through the training data as you want/deem necessary: a. Take a record, add noise using a random number generator, to (some/all/one?) predictor variable, based on the mean and s.d. b. train on this imaginary record Now, I am not too sure that this is "statistically" acceptable, and I haven't tried it myself, but it seems to me that this would cause the nn to train on a more diverse range of data. Thus, whatever system of weights it came up with would be better able to handle the unknown data coming at it in the test data. It also might be a form of "smoothing" in the nn weight space. Alas, I can't remember where I saw this idea. Nor do I remember the results. Half the time, for me, implementation in code takes up more time than the experiment. So I always look with trepidation on new ideas. It's the old "up to your ass in alligators" problem. So, has anybody done this , or read any reports on anybody doing this ? Please let me know. Thanks, Phil Neal phil@iris.iphc.washington.edu ------------------------------ Subject: Neural Nets in Autonomous Land Vehicles From: kagan@computervision.bristol.ac.uk Date: Fri, 05 Apr 91 15:53:33 +0100 I am looking for all kinds of information, software etc. about the neural network applications in autonomous land vehicles. I am particularly interested in steering and velocity control of the vehicle. Thanks in advance. Kagan Ozerhan <K.Ozerhan@bristol.ac.uk> University of Bristol University Walk Queens Building Room 1.65 Bristol BS8 1TR U.K. ------------------------------ Subject: Proceedings of Third NN and PDP Conference, Indiana-Purdue From: SAYEGH@CVAX.IPFW.INDIANA.EDU Date: Wed, 10 Apr 91 20:41:22 -0400 The proceedings of the THIRD conference on Neural Networks and Parallel Distributed Processing held in April 1990 at Indiana-Purdue University in Ft Wayne can be obtained by writing to: Ms. Sandra Fisher Physics Department Indiana University-Purdue University Ft Wayne, IN 46805 and including $5 + $1 for mailing and handling. Checks should be made payable to The Indiana-Purdue Foundation. The 109 page proceedings contain the following articles: INTEGRATED AUTONOMOUS NAVIGATION BY ADAPTIVE NEURAL NETWORKS Dean A. Pomerleau Department of Computer Science Carnegie Mellon University APPLYING A HOPFIELD-STYLE NETWORK TO DEGRADED PRINTED TEXT RESTORATION Arun Jagota Department of Computer Science State University of New York at Buffalo RECENT STUDIES WITH PARALLEL, SELF-ORGANIZ- ING, HIERARCHICAL NEURAL NETWORKS O.K. Ersoy & D. Hong School of Electrical Engineering Purdue University INEQUALITIES, PERCEPTRONS AND ROBOTIC PATH- PLANNING Samir I. Sayegh Department of Physics Indiana University-Purdue University GENETIC ALGORITHMS FOR FEATURE SELECTION FOR COUNTERPROPAGATION NETWORKS F.Z. Brill & W.N. Martin Department of Computer Science University of Virginia MULTI-SCALE VISION-BASED NAVIGATION ON DIS- TRIBUTED-MEMORY MIMD COMPUTERS A.W. Ho & G.C. Fox Caltech Concurrent Computation Program California Institute of Technology A NEURAL NETWORK WHICH ENABLES SPECIFICATION OF PRODUCTION RULES N. Liu & K.J. Cios The University of Toledo PIECE-WISE LINEAR ESTIMATION OF MECHANICAL PROPERTIES OF MATERIALS WITH NEURAL NETWORKS I.H. Shin, K.J. Cios, A. Vary* & H.E. Kautz* The University of Toledo & NASA Lewis Re- search Center* INFLUENCE OF THE COLUMN STRUCTURE ON INTRA- CORTICAL LONG RANGE INTERACTIONS E. Niebur & F. Worgotter California Institute of Technology LEARNING BY GRADIENT DESCENT IN FUNCTION SPACE Ganesh Mani University of Wisconsin-Madison REAL TIME DYNAMIC RECOGNITION OF SPATIAL TEMPORAL PATTERNS M. F. Tenorio School of Electrical Engineering Purdue University A NEURAL ARCHITECTURE FOR COGNITIVE MAPS Martin Sonntag Cognitive Science & Machine Intelligence Lab University of Michigan P.S. The Fourth Conference is scheduled to start April 11, 91 at 6pm in the Classroom Medical Building, CM159, of the Fort Wayne Campus of Indiana and Purdue. A previous announcement of this conference was made on the list. ------------------------------ Subject: Neural Compuation Vol 3 Issue 1 From: Terry Sejnowski <tsejnowski@UCSD.EDU> Date: Sat, 13 Apr 91 22:57:22 -0700 NEURAL COMPUTATION - Volume 3 Issue 1 - Spring 1991 Review: Deciphering the Brain's Codes Masakazu Konishi Letters: Synchronization of Bursting Action Potential Discharge in a Model Network of Neocortical Neurons Paul Bush and Rodney Douglas Parallel Activation of Memories in an Oscillatory Neural Network D. Horn and M. Usher Organization of Binocular Pathways: Modeling and Data Related to Rivalry Sidney R. Lehky Dynamics and Formation of Self-Organizing Maps Jun Zhang A Method for Reducing Computation in Networks with Separable Radial Basis Functions Terrence D. Sanger Adaptive Mixtures of Local Experts Robert A. Jacobs, Michael I. Jordan, Steven J. Nowlan, and Geoffrey E. Hinton Efficient Training of Artificial Neural Networks for Autonomous Navigation Dean A. Pomerleau Sequence Manipulation Using Parallel Mapping David S. Touretzky and Deirdre W. Wheeler Parsing Complex Sentences with Structured Connectionist Networks Ajay N. Jain Rules and Variables in Neural Nets Venkat Ajjanagadde and Lokendra Shastri TAG: A Neural Network Model for Large-Scale Optical Implementation Hyuek-Jae Lee, Soo-Young Lee, and Sang-Yung Shin SUBSCRIPTIONS - VOLUME 3 ______ $35 Student ______ $55 Individual ______ $110 Institution Add $18. for postage and handling outside USA (Back issues are available for $28 each.) MIT Press Journals, 55 Hayward Street, Cambridge, MA 02142. (617) 253-2889. MasterCard and VISA accepted ------------------------------ Subject: TR available: Catastrophic forgetting From: Bob French <french@cogsci.indiana.edu> Date: Fri, 29 Mar 91 22:03:45 -0500 The following brief technical report is available from the Center for Research on Concepts and Cognition at Indiana University: USING SEMI-DISTRIBUTED REPRESENTATIONS TO OVERCOME CATASTROPHIC FORGETTING IN CONNECTIONIST NETWORKS Robert M. French Center for Research on Concepts and Cognition Indiana University 510 North Fess Bloomington, IN 47408 e-mail: french@cogsci.indiana.edu In connectionist networks, newly-learned information rapidly destroys previously-learned information unless the network is continually retrained on the old information. This behavior, known as catastrophic forgetting, is unacceptable both for practical purposes and as a model of mind. This paper advances the claim that catastrophic forgetting is a direct consequence of the overlap distributed representations and can be reduced by reducing this overlap. It is also suggested that there is an inevitable trade-off between generalization and forgetting. A simple algorithm is presented that allows a standard feedforward backpropagation network to develop "semi-distributed representations", thereby significantly reducing the problem of catastrophic forgetting. TO OBTAIN A COPY OF THIS PAPER: unix>ftp cogsci.indiana.edu (or ftp 129.79.238.6) ftp>user: anonymous ftp>passwd: ident ftp>cd pub ftp>binary ftp>get french.forgetting.ps.Z ftp>quit unix>uncompress french.forgetting.ps.Z unix>lpr -P(your laser printer) french.forgetting.ps Hard copies may be requested by sending e-mail to: french@cogsci.indiana.edu or by writing directly to C.R.C.C. at the address indicated above. ------------------------------ Subject: prepint by Lumer & Huberman: "Binding Hierarchies:..." From: Andreas Weigend <andreas%psych@Forsythe.Stanford.EDU> Date: Wed, 10 Apr 91 21:48:16 -0700 The following preprint is available in hardcopy form. It can be obtained by sending e-mail to: lumer@parc.xerox.com Please do NOT reply to this message. ______________________________________________________________________ Binding Hierarchies: A Basis for Dynamic Perceptual Grouping E. Lumer and B. A. Huberman Stanford University and Xerox PARC Abstract Since it has been suggested that the brain binds its fragmentary representations of perceptual events via phase-locking of stimulated neuron oscillators, it is important to determine how extended synchronization can occur in a clustered organization of cells possessing a distribution of firing rates. In order to answer that question, we establish the basic conditions for the existence of a binding mechanism based on phase-locked oscillations. In addition, we present a simple hierarchical architecture of feedback units which not only induces robust synchronization within and segregation between perceptual groups, but also serves as a generic binding machine. ______________________________________________________________________ ------------------------------ Subject: Neural Network Seminar From: noordewi@cs.RUTGERS.EDU Date: Fri, 12 Apr 91 13:00:24 -0400 RUTGERS UNIVERSITY Dept. of Computer Science/Dept. of Mathematics Neural Networks Colloquium Series --- Spring 1991 C. L. Giles NEC Research Institute Teaching Recurrent Neural Networks to be Finite State Machines (Digraphs) Abstract Recurrent neural networks are natural models for encoding and learning temporal sequences. If these temporal sequences are strings from the languages of formal grammars, then teaching a neural network to learn these sequences is a form of grammatical inference. We demonstrate how to train second-order recurrent networks with real-time learning algorithms to be finite state machines. In particular, we present extensive simulation results which show that simple regular grammars are easy to learn. We devise and use heuristic clustering algorithms which extract finite state machines or digraphs from recurrent neural networks during and after training. The resultant finite state machines usually have large numbers of states and can be reduced in complexity to minimal finite state machines using a standard minimization algorithm. Depending on the training method and type of training set, different minimal finite state machines emerge. If the grammar is well learned, then identical finite state machines are produced in the minimization process. These finite state machines constitute an equivalence class of neural networks which covers different numbers of neurons and different initial conditions. This can be interpreted as a measure of how well a set of strings and its generative grammar are learned. We present a video of the learning process and show the emergent finite state machines during and after training. April 17, 1991 Busch Campus --- 4:30 p.m., room 217 SEC host: Mick Noordewier (201/932-3698) finger noordewi@cs.rutgers.edu for further schedule information [[ Editor's Note: Ooops, this went out AFTER the talk. Sorry about that. -PM ]] ------------------------------ Subject: Neural Nets Workshop (IWANN 91) From: Ignacio Bellido Montes <ibm@dit.upm.es> Date: Mon, 08 Apr 91 13:08:01 +0200 [[ Editor's Note: In the interest of space, I deleted the LaTeX version of the announcement. -PM ]] There will be a Neural Nets Workshop Granada (Spain) next September. I think this is the first international workshop of this kind that will be held in Spain, I hope all of you can send contributions and participate. Here I send you the electronic version of the CFP followed by the LaTeX version. If you have any question about the workshop, please ask me or the people of the workshop secretary. I'm not on the organization but I can help finding people etc... About Granada, I can tell you this is one of the most wonderful cities in the Spain and ... in the World. It has one of the most famous and wonder arab contributions to the culture, "La Alhambra", and small and quiet lanes on the old side of the city, the "Albaicin", that transmit the visitor a very good feeling. I hope you be able to participate and enjoy... See you in Granada. Gregorio Fernandez Dpt. Ingenieria de Sistemas Telematicos ETSI Telecomunicacion- UPM Ciudad Universitaria 28040 Madrid, Spain E-mail: gfernandez@dit.upm.es =----------------------------------------------------------------------------- INTERNATIONAL WORKSHOP ON ARTIFICIAL NEURAL NETWORKS IWANN'91 First Announcement and Call for Papers Granada, Spain September 17-19, 1991 ORGANISED AND SPONSORED BY Spanish Chapter of the Computer Society of the IEEE, AEIA (IEEE Affiliate Society), and Department of Electronic and Computer Technology. University of Granada. Spain. SCOPE Artificial Neural Networks (ANN) were first developed as structural or functional modelling systems of natural ones, featuring the ability to perform problem-solving tasks. They can be thought as computing arrays consisting of series of repetitive uniform processors (neuron-like elements) placed on a grid. Learning is achieved by changing the interconnections between these processing elements. Hence, these systems are also called connectionist models. ANN has become a subject of wide-spread interest: they offer an odd scheme-based programming standpoint and exhibit higher computing speeds than conventional von-Neumann architectures, thus easing or even enabling handling complex task such as artificial vision, speech recognition, information recovery in noisy environments or general pattern recognition. In ANN systems, collective information management is achieved by means of parallel operation of neuron-like elements, into which information processing is distributed. It is intended to exploit this highly parallel processing capability as far as possible in complex problem-solving tasks. Cross-fertilization between the domains of artificial and real neural nets is desirable. The more genuine problems of biological computation and information processing in the nervous system still remain open and contributions in this line are more than welcome. Methodology, theoretical frames, structural and organizational principles in neuroscience, self-organizing and co- operative processes and knowledge based descriptions of neural tissue are relevant topics to bridge the gap between the artificial and natural perspectives. The workshop intends to serve as a meeting place for engineers and scientists working in this area, so that present contacts and relationships can be further increased. The workshop will comprise two complementary activities: . scientific and technical conferences, and . scientific communications sessions. TOPICS The workshop is open to all aspects of artificial neural networks, including: 1. Neural network theories. Neural models. 2. Biological perspectives 3. Neural network architectures and algorithms. 4. Software developments and tools. 5. Hardware implementations 6. Applications. LOCATION Facultad de Ciencias Campus Universitario de Fuentenueva Universidad de Granada 18071 GRANADA. (SPAIN) LANGUAGES English and Spanish will be the official working languages. English is preferable as the working language. CALL FOR PAPERS The Programme Committee seeks original papers on the six above mentioned areas. Survey papers on the various available approaches or particular application domains are also sought. In their submitted papers, authors should pay particular attention to explaining the theoretical and technical choices involved, to make clear the limitations encountered and to describe the current state of development of their work. INSTRUCTIONS TO AUTHORS Three copies of submitted papers (not exceeding 8 pages in 21x29.7 cms (DIN-A4), with 1,6 cm. left, right, top and bottom margins) should be received by the Programme Chairman at the address below before June 20, 1991. The headlines should be centered and include: . the title of paper in capitals . the name(s) of author(s) . the address(es) of author(s), and . a 10 line abstract. Three blank lines should be left between each of the above items, and four between the headlines and the body of the paper, written in English, single-spaced and not exceeding the 8 pages limit. All papers received will be refereed by the Programme Committee. The Committee will communicate their decision to the authors on July 10. Accepted papers will be published in the proceedings to be distributed to workshop participants. In addition to the paper, one sheet should be attached including the following information: . the title of the paper, . the name(s) of author(s), . a list of five keywords, . a reference to which of the six topics the paper concerns, and . postal address of one of the authors, with phone and fax numbers, and E-mail (if available). We intend to get in touch with various international publishers (such as Springer-Verlag and Prentice-Hall) for the final version of the proceedings. Contributions to be sent to: Prof. Jose Mira Dpto. Informatica y Automatica UNED C/Senda del Rey s/n 28040 MADRID (Spain) Tel. (34) 1 5 44 60 00 Fax: (34) 1 5 44 67 37 ACCOMMODATION A list of available Hotels will be sent on registration. Hotel reservations can be made directly by each participant with the local agency below. All request should be addressed to: Viajes Internacional Expreso (VIE) Galerias Preciados Carrera del Genil, s/n 18005 GRANADA (Spain) Tel. (34) 58-22.44.95, -22.75.86, -224944 Telex: 78525 We can only guarantee to accept reservation received by July 25. REGISTRATION FEE . Regular fee: 35.000 ptas. . IEEE, AEIA and ATI members fee: 30.000 ptas. . Scholarship holders fee: 5.000 ptas. Inscription payments: Transfer to: IWANN'91 account number: 16.142.512 Caja Postal (Code: 2088-2037.1) Camino de Ronda, 138 18003 GRANADA (SPAIN) or alternatively, cheque made out to: IWANN'91 (16.142.512) Secretariat address: Departamento de Electronica y Tecnologia de Computadores Facultad de Ciencias Universidad de Granada 18071 GRANADA (SPAIN) FAX: 34-58-24.32.30 or 34-58-27.42.58 Phone: 34-58-24.32.26 E-Mail: jmerelo@ugr.es aprieto@ugr.es PROGRAM AND ORGANIZATION COMMITTEE Senen Barro Univ. de Santiago Joan Cabestany Univ. Pltca. de Catalunya Jose Antonio Corrales Univ. Oviedo. Gregorio Fernandez Univ. Pltca. de Madrid J. Simoes da Fonseca Univ. de Lisboa Antonio Lloris Univ. Granada Javier Lopez Aligue Univ. de Extremadura. Jose Mira (Programme Chairman) UNED. Madrid Roberto Moreno Univ Las Palmas Gran Canaria Alberto Prieto (Organization Chairman) Univ. Granada Francisco Sandoval Univ. de Malaga Carmen Torras Instituto de Cibernetica. CSIC. Barcelona Elena Valderrama CNM- Univ. Autonoma de Barcelona LOCAL ORGANIZING COMMITTEE (Universidad de Granada) Juan Julian Merelo Julio Ortega Francisco J. Pelayo Begona del Pino Alberto Prieto (To be completed and returned as soon as possible to: Departamento de Electronica. Facultad de Ciencias. Univ. de Granada. 18071 GRANADA (SPAIN). FAX (34)-58-24.32.30) Block letters, please Name: Company/Organization: Address: State/Country: E-mail: Phone: Fax: Please tick as appropriate: I intend to: attend to the workshop submit a paper Name(s) of Author(s): Provisional Title: ------------------------------ End of Neuron Digest [Volume 7 Issue 21] ****************************************